作者位次 #6/8ICLR 2026Poster· Deep Learning / Theory
Fast Catch-Up, Late Switching: Optimal Batch Size Scheduling via Functional Scaling Laws
Jinbo Wang · Binghui Li · Zhanpeng Zhou · Mingze Wang · yuxuan sun · Jiaqi Zhang · Xunliang Cai · Lei Wu
首次获取该候选人的完整档案及 StartLux 评估共 20 积分;每位用户单独解锁,之后查看免费。主动刷新 20 积分。
实时查找电子邮箱,费用不超过20积分;未找到邮箱不扣费。
Jinbo Wang · Binghui Li · Zhanpeng Zhou · Mingze Wang · yuxuan sun · Jiaqi Zhang · Xunliang Cai · Lei Wu
Yuxuan Hu · Jianchao Tan · Jiaqi Zhang · Wen Zan · Pingwei Sun · Yifan Lu · Xunliang Cai · Jing Zhang
In this work, we conduct a systematic analysis of Native Sparse Attention (NSA) and propose targeted improvements that enhance long-context modeling. A key insight is that alternating between local (sliding-window) and global (compression/selective) attention across layers, rather than using fixe...